Allow the `Self` type to be used in impls.

[nrc]

Closes #310

See also some discussion in discuss and discussion in the Rust issue tracker.

[nrc]

rendered

[nrc]

And an implementation to show how easy it is - less than 100 lines of code changed (including the test) and no changes outside of libsyntax.

[aturon]

Nice! I've often wanted this.

[sinistersnare]

+1 simple and useful for macros

[pythonesque]

+1, always seemed like an oversight that you couldn't do this.

[nikomatsakis]

It'd certainly be convenient and I've kind of expected we'll add it sooner or later. As far as the implementation goes, I would have expected to handle it in astconv, but I guess doing it in libsyntax is ok too and avoids needing to modify resolve, though you'll get some surprising errors (e.g., if you have a resolve error in the self type on the impl).

[nrc]

I realised today that if we want to accommodate Self::T paths to associated types we would need to either do this in resolve or add the facility to write a path to an associated type via a concrete type (which I think we want to add, but not any time soon). Given the latter is coming, I would prefer to land this and avoid doing the expansion in paths with more than one segment.

[nikomatsakis]

On Mon, Dec 15, 2014 at 05:27:54PM -0800, Nick Cameron wrote:

add the facility to write a path to an associated type via a
concrete type (which I think we want to add, but not any time soon)

I expect this to fall out of the refactoring for AT we've been
discussing.

[nrc]

This RFC has been accepted, r=nikomatsakis

Tracking issue

Small, easy to implement, and widely desired. This will NOT block 1.0.